User Guide

Caveman prompt mode

On long chats and bulk runs the output cost adds up, and most of those words are filler. Caveman is a prompt style that strips articles, padding and pleasantries from Claude's replies — fewer tokens, lower cost, quicker to read.

What you get

Output cost drops noticeably on long chats and bulk runs.
Replies are faster to read — no boilerplate to skim past.
Code, errors and structured data stay exact, so nothing technically relevant is compressed away.

Caveman is a community style by Julius Brussee (github.com/JuliusBrussee/caveman). It comes in several levels, from a light trim to an aggressive one, with full as the sensible default.

How you use it

Open Settings → Caveman.
Toggle Caveman On to enable it for this project.
Pick a level — the selector appears once the switch is on.
Save. From now on the task chat (and any Claude session this project starts) replies in the chosen style.
Disable any time — your original prompts are untouched, only the trimming instruction is dropped.

Caveman applies to the inline task chat, and pairs naturally with cost tracking when you're tuning spend.

Settings Integrations tab with the Caveman mode toggle — Settings → Integrations — enable Caveman mode for token-tight runs.